Received-SPF: pass (google.com: domain of linux-nfs-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Content-Type: text/plain; charset="utf-8"
Content-Transfer-Encoding: quoted-printable
MIME-Version: 1.0
From:   "NeilBrown" <neilb@suse.de>
To:     "Trond Myklebust" <trondmy@hammerspace.com>
Cc:     "chuck.lever@oracle.com" <chuck.lever@oracle.com>,
        "linux-nfs@vger.kernel.org" <linux-nfs@vger.kernel.org>
Subject: Re: Question about RDMA server...
In-reply-to: <50ea0e392fd8b0b1bec987e5c5f923c7baef439f.camel@hammerspace.com>
References: <82662b7190f26fb304eb0ab1bb04279072439d4e.camel@hammerspace.com>,
 <1114899D-BBF5-4CB1-9126-E4E652ACAAB6@oracle.com>,
 <5DCBD9EB-7721-48FC-9EBD-58B7DF05A704@oracle.com>,
 <8af942181abb39cd7ce8fe91be9c4c2f8c9f2c56.camel@hammerspace.com>,
 <2E4807E4-5086-4F15-B790-8D952B394FE5@oracle.com>,
 <974fa169124661c2ce5ed549d499837435cc7b4c.camel@hammerspace.com>,
 <E7FD566B-0570-4D14-9936-5C737B619E0B@oracle.com>,
 <b9b98ad9b21f228566a5ebd643198c669c9f3408.camel@hammerspace.com>,
 <930A468B-E674-4F5D-8BC0-DB9F45611A9D@oracle.com>,
 <50ea0e392fd8b0b1bec987e5c5f923c7baef439f.camel@hammerspace.com>
Date:   Tue, 05 Apr 2022 15:00:06 +1000
Message-id: <164913480601.10985.1519370018759325971@noble.neil.brown.name>
Precedence: bulk

On Tue, 05 Apr 2022, Trond Myklebust wrote:
> On Mon, 2022-04-04 at 15:34 +0000, Chuck Lever III wrote:
> >=20
....
> >=20
> > I gotta ask: Why does cache_defer_req() check if thread_wait is zero?
> > Is there some code in the kernel now that sets it to zero? I see only
> > these two values:
> >=20
> > =C2=A0784=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 if (!test_bit(S=
P_CONGESTED, &pool->sp_flags))
> > =C2=A0785=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 rqstp->rq_chandle.thread_wait =3D 5*HZ;
> > =C2=A0786=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 else
> > =C2=A0787=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=
=A0=C2=A0=C2=A0=C2=A0=C2=A0=C2=A0 rqstp->rq_chandle.thread_wait =3D 1*HZ;
>=20
>=20
> That test goes all the way back to commit f16b6e8d838b ("sunrpc/cache:
> allow threads to block while waiting for cache update."), so probably
> more of a question for Neil.

12 years ago....  the commit message suggests that the "underlying
transport" might not approve of sleeping, but I cannot think what that
might mean.  Probably it seemed to mean something earlier in the design
process but I discarded the idea without discarding the code.

> I don't see any immediate problems with removing it: even if someone
> does set a value of 0, the wait code itself should cope.
>=20
Agreed.

NeilBrown